Hardware

Three form factors.
One mission.

From a quiet desk-side workstation to a data center powerhouse. Own it once, run unlimited AI forever.

Book a Demo View Specs
Compact AI Desktop Workstation

ZeroBoxx Core

No rack required. Desk-side or AV cabinet deployment for SMBs, clinics, and law firms.

Book a Demo
Compute
GPU 2× NVIDIA B80 PCIe Blackwell Architecture
VRAM 160 GB total 80 GB per card
CPU Dual Intel Xeon Platinum 6 High-core-count workstation CPU
System RAM 512 GB DDR5 ECC Error-correcting registered DIMMs
Storage
NVMe 4× 3.84 TB PCIe Gen5 SSD, ~15 TB usable
Connectivity
Network Dual 10/25 GbE SFP28, RJ45 management port
Software
OS Ubuntu 24.04 LTS Pre-installed
AI Stack NVIDIA AI Dev Tools Ollama, CUDA, cuDNN
API OpenAI-compatible REST API included
Powered by NVIDIA Blackwell
Most Powerful
AI Rack Server (8U)

ZeroBoxx Max

Organization-wide AI. Multiple frontier models running in parallel across your enterprise.

Book a Demo
Compute
GPU 8× NVIDIA H200 SXM5 Hopper Architecture, NVLink fabric
VRAM 1.1 TB total 141 GB HBM3e per GPU, unified via NVLink
CPU Dual Intel Xeon Platinum 8592+ 64-core flagship server CPU
System RAM 4 TB DDR5 ECC Registered DIMMs
Storage
NVMe 16× 7.68 TB RAID 6 PCIe Gen5 SSD, ~92 TB usable
Connectivity
Network 4× 400 GbE / InfiniBand HDR Multi-rail fabric
Management BMC / IPMI Out-of-band management
Software
OS Ubuntu 24.04 LTS Pre-installed
AI Stack NVIDIA AI Dev Tools Ollama, CUDA, cuDNN, TensorRT
API OpenAI-compatible REST API, gRPC, Python SDK
Powered by NVIDIA Blackwell

Included with every ZeroBoxx

Air-Gapped by Default

Zero data sent to cloud. Ever.

Unlimited Inference

No rate limits or per-token charges.

OpenAI-Compatible API

Drop-in replacement. Change one URL.

Any Open Model

Llama, Mistral, DeepSeek and more.

Full Fine-Tuning

Train on your own data, on your hardware.

Day-One Ready

Ships pre-configured, running in hours.

Hot-Swap Models

Change models without rebooting.

LAN-Only Deployment

No internet required. Ever.

Frequently asked questions

What AI models can I run on ZeroBoxx?

Any open-weight model. Llama 3.x, Mistral, Mixtral, DeepSeek, Qwen, and others are all supported via Ollama or vLLM. If a model fits in your VRAM pool, it runs.

How long does setup take?

Under a day. ZeroBoxx ships pre-configured with Ubuntu 24.04 LTS, NVIDIA drivers, CUDA, Ollama, and an OpenAI-compatible API. Most teams are running inference within hours of delivery.

Can I fine-tune models on ZeroBoxx?

Yes. The VRAM headroom on ZeroBoxx Pro and Max is specifically designed for LoRA and full fine-tuning workflows on your own proprietary data.

Does ZeroBoxx require internet access?

No. All inference runs entirely on-premise. Internet is only needed for the initial setup and optional model downloads. After that, ZeroBoxx operates completely air-gapped.

What support is included?

Standard support is included with every ZeroBoxx. Enterprise SLAs and on-site support contracts are available for Pro and Max customers.

DataAegis: Knowledge Layer

Your data, your database.

DataAegis connects your knowledge bases and enterprise databases to the AI OS layer and keeps everything inside your network. RAG, vector search, and database connectors without cloud exposure.

Enterprise Security

Built for regulated industries.

Enterprise-grade security is baked into the AI OS layer, not an add-on.

HIPAASOC 2GDPRFedRAMP-gradeAir-Gap ReadyZero-Trust

Ready to own your AI?

Book a demo to see ZeroBoxx in action and get a configuration recommendation for your team.

Book a Demo